What Drives Electric Vehicle Diffusion?: An Agent-Based Approach to Assess Factors and Market Effects in Norway

(1)

IN THE FIELD OF TECHNOLOGY DEGREE PROJECT

DESIGN AND PRODUCT REALISATION AND THE MAIN FIELD OF STUDY INDUSTRIAL MANAGEMENT, SECOND CYCLE, 30 CREDITS STOCKHOLM SWEDEN 2017,

What Drives Electric Vehicle Diffusion?

An Agent-Based Approach to Assess Factors and Market Effects in Norway

ALBIN GROPP

FREDRIK OHLSSON

(2)

(3)

What Drives Electric Vehicle Diffusion?

An Agent-Based Approach to Assess Factors and Market Effects in Norway

Albin Gropp

Fredrik Ohlsson

(4)

Vad påverkar diffusionen av elektriska fordon?

En agentbaserad modell för att bedöma faktorer och marknadseffekter i Norge

Albin Gropp Fredrik Ohlsson

Examensarbete INDEK 2017:42 KTH Industriell teknik och management

Industriell ekonomi och organisation SE-100 44 STOCKHOLM

(5)

Master of Science Thesis INDEK 2017:42

What Drives Electric Vehicle Diffusion?

An Agent-Based Approach to Assess Factors and Market Effects in Norway

Albin Gropp Fredrik Ohlsson

Approved Examiner Supervisor

2017-06-07 Cali Nuur Niklas Arvidsson

Commissioner Contact person

Fortum Sverige AB Catarina Naucler

Abstract

Environmental sustainability and energy efficiency have over the previous years become important topics on the political agenda to reduce emissions and prevent global warming. Due to the demand for sustainable transportation, the interest in electric vehicles (EVs) has increased as the functionality is coming closer to par with the incumbent internal combustion engine vehicles.

Norway is the frontrunner in EV adoption, why dynamics of the Norwegian car market have undergone a shift. In addition, EVs are now entering the used car market where the presence previously has been limited. Understanding the factors that affect the decisions to purchase an EV and thereby the diffusion, is non-trivial as technology advancements, political measures and consumer preferences are continuously changing. Stakeholders are closely monitoring the developments on the Norwegian market in order to gain insight on products, infrastructure investments, and measures to deal with the change.

This study investigates the diffusion factors for EV adoption in Norway by using an agent-based model to simulate the consumer behavior in different scenarios spanning until 2030. The model is initialized using data from 3962 individual responses in a Norwegian survey. Further, how the used car market is affected by the introduction of EVs is investigated using data from Norway’s largest online marketplace for used vehicles.

The results from the simulations show that EV diffusion are positively affected by combining political incentives and technology developments, while the perception of EVs created by social interactions between consumers can expedite adoption even further. If such factors are combined, the simulations show that 92% of the Norwegian fleet could be electric in 2030, coupled with low emissions. Moreover, the analysis of the used car market show that gasoline and diesel cars have experienced increased depreciation over recent years. The findings can be used by stakeholders as a guideline toward where the market is heading, and further research should include tailoring survey questions to fit the model as well as more extensive analysis of value retention for specific car models.

Keywords: electric vehicles, diffusion of innovations, agent-based model, market development,

(6)

(7)

Examensarbete INDEK 2017:42

Vad påverkar diffusionen av elektriska fordon?

En agentbaserad modell för att bedöma faktorer och marknadseffekter i Norge

Albin Gropp Fredrik Ohlsson

Godkänt Examinator Handledare

2017-06-07 Cali Nuur Niklas Arvidsson

Uppdragsgivare Kontaktperson

Fortum Sverige AB Catarina Naucler

Sammanfattning

Hållbar utveckling och energieffektivitet har under de senaste åren blivit viktiga ämnen på den politiska agendan i syfte att minska utsläpp och förhindra global uppvärmning. Till följd av efterfrågan på miljövänlig transport har intresset för elbilar ökat kraftigt, samtidigt som funk- tionaliteten närmar sig den för de traditionella fordonen med förbränningsmotor. Norge är föregångare i att gå över till elbilar, varför dynamiken på den norska bilmarknaden har genomgått ett skifte. I tillägg har elbilar nu börjat ta plats på andrahandsmarknaden, där de tidigare varit nästintill obefintliga. Att förstå de faktorer som påverkar beslutprocessen att köpa elbil och därmed diffusionen är inte trivialt då teknisk utveckling, politiska åtgärder och konsumenters preferenser ständigt är i förändring. Intressenter bevakar denna utveckling ingående för att få in- sikt om produkter, infrastrukturinvesteringar och åtgärder för att kunna hantera teknologiskiftet.

Det här examensarbetet undersöker faktorerna som påverkar diffusionen av elbilar i Norge, genom att använda en agentbaserad modell för att simulera konsumenternas beteenden i olika scenarier fram till 2030. Modellen initieras med data från 3962 individuella svar från en norsk enkätstudie.

Vidare undersöks hur andrahandsmarknaden påverkas av att fler elbilar introduceras genom analys av data från Norges största marknadsplatser för begagnade fordon.

Resultaten från simuleringarna visar att diffusionen av elbilar påverkas positivt genom kombi- nationen av politiska incitament och teknisk utveckling, samtidigt som bilden av elbilar skapad av social interaktion mellan konsumenter kan påskynda skiftet än mer. När dessa faktorer kombineras, visar simuleringarna att 92% av Norges bilflotta kan vara elektrisk år 2030, resul- terande i låga utsläpp av växthusgaser. Analysen av andrahandsmarknaden visar att bensin- och dieselbilar har upplevt ökade värdeförluster under de senaste åren. Resulaten kan användas av intressenter som indikation på vart marknaden är på väg. Framtida arbete inkluderar skräd- darsydda enkätfrågor för modellen samt en djupgående analys av värdeminskningen på specifika bilmodeller.

Nyckelord: elbilar, diffusion, agentbaserad modell, marknadsutveckling, andrahandsmarknad

(8)

(9)

List of Figures

1 Source of purchased vehicles for survey participants (Figenbaum & Kolbenstvedt,

2016) . . . 9

2 Rogers diffusion of innovation. The bell-shaped frequency curve represent the number of individuals that adopt to a new technology per time unit, and the S-shaped curve is the cumulative basis (Rogers, 1962;2003) . . . 12

3 General structure of an agent-based model (van Dam et al., 2012) . . . 14

4 Consumat framework overview (Jager, 2000) . . . 19

5 Abstract overview of the STECCAR model (Kangur, 2014) . . . 21

6 Research process and work flow throughout the thesis . . . 27

7 Overview of scenario dimensions; Control, Policy Alleviation, Technology Devel- opment and Combination. The stratisfied sample represents the Social dimension 31 8 Diffusion of EVs in the Control scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 34

9 Diffusion of EVs in the Emissions Tax scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 35

10 Diffusion of EVs in the Emissions Tax:BEV scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 36

11 Diffusion of EVs in the Battery scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation . . . 37

12 Diffusion of EVs in the Fast Charge Probability scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 38

13 Diffusion of EVs in the Fuel Cost scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 39

14 Diffusion of EVs in the Combination Fast Charge scenario over the years 2016- 2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 40

15 Diffusion of EVs in the Combination All scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 41

16 Diffusion of EVs in the Combination All:BEV scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 42

17 Diffusion of EVs in the Control Social scenario over the years 2016-2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 43

18 Diffusion of EVs in the Combination ALL:BEV Social scenario over the years 2016- 2030. The solid lines represent the average value, and the faded area standard deviation at the point in time . . . 44

19 Diffusion in the different scenarios sorted on BEV diffusion in the end of the simulation, year 2030 . . . 45 20 Average emissions [gCO2/km] with standard deviation error bands for the overall

(14)

21 Fraction of vehicles sold on the used car market by fuel type in the Control scenario 47 22 Total number of registered cars by fuel type (Derived from data by SSB, 2017) . 47 23 Number of used car ads on FINN, 2011-2016 . . . 48 24 Median price development of used cars by fuel type and year model on FINN.

Year models 2011-2013 . . . 49 25 Median price development of used cars by fuel type and year model on FINN.

Year models 2014-2016 . . . 49 26 First year depreciation of cars by fuel type and year model. Numbers for year

2017 and forward are linearly forecasted together with logarithmic trend . . . . 50 27 Average ownership duration over the simulation run. . . 55

x

(15)

List of Tables

1 Parameters used in STECCAR (DS=Dataset; E=Estimated; I=Independent) . . 26 3 Overview of all scenarios showing in which dimension adjustments are made . . . 32 4 Scenario parameters . . . 33 5 Emission tax categories with corresponding tax level . . . 35

(16)

(17)

Acronyms

ABM Agent-Based Model

BEV Battery Electric Vehicle

COMPETT Competitive Electric Town Transport

EV Electric Vehicle

HEV Hybrid Electric Vehicle

ICE Internal Combustion Engine (Vehicle) PHEV Plug-in Hybrid Electric Vehicle

STECCAR Simulating the Transition to Electric Cars using the Consumat Agent Rationale TCO Total Cost of Ownership

VAT Value Added Tax

YoY Year-on-Year

(18)

(19)

Foreword and Acknowledgements

This study is the derivative work of a master thesis conducted at the department of Industrial Economics and Management at KTH Royal Institute of Technology, Stockholm. The study was carried out during the spring 2017 in collaboration with Fortum Sverige.

(Part of) the data used here are taken from "COMPETT - Konkurransdyktig elektrisk transport i byer, 2014". The data is collected by Erik Figenbaum, Transportokonomisk Institutt. The study is funded by the Norwegian Research Council. The data is formated to be anonymous by NSD - Norsk Senter for Research Data AS. Neither Erik Figenbaum, Institute of Transport Economics, Research Council of Norway and NSD are responsible for the analysis of the data or interpretations that are made here in this study.

We would like to thank everyone at Fortum for making us feel welcomed. Whenever help was needed it was there! A special thanks goes out to Catarina Naucler and Staffan Sandblom for the interesting discussions and being so supportive. We are very thankful for this exciting opportunity you have given us. To our supervisor at KTH Royal Institute of Technology, Niklas Arvidsson, thank you for your guidance and support with the thesis. Also, a thanks to our peers for giving us valuable feedback during the seminars.

Lastly, we would like to express our gratitude to friends and family for supporting us through these academic years leading up to this culmination point of our studies.

Albin Gropp and Fredrik Ohlsson June 5, 2017

Stockholm, Sweden

(20)

(21)

1 Introduction

This chapter presents the background for the problematization along with the purpose of the study. The purpose is followed by the research questions, contribution, delimitations and limitations, and the thesis structure.

1.1 Background

Increasing demand of sustainable solutions for energy production, energy consumption, and transportation has in the few previous years led to an increased interest in electric vehicles (EVs).

Nevertheless, EVs have previously been characterized by low functionality and high cost, which can be considered a consequence of the dominance of internal combustion engine (ICE) vehicles , the oil industry, transport networks, technologies, and institutions with which it has co-evolved (Struben & Sterman, 2008). This have in addition with the size of the automotive industry created large positive feedback loops and path dependencies that have provided advantage to the incumbent ICE technology. However, factors such as technological advancements and political incentives for clean transportation have over recent years improved EV sales despite the barriers to EV adoption, including; high initial cost, range anxiety, and scarcity of charging infrastructure among others (Adepetu et al., 2016; Boulanger et al., 2011; Struben & Sterman, 2008).

In the fourth quarter of 2015, the global cumulative number of EVs on the roads passed 1 million cars and is steadily increasing (IEA, 2016; BNEF, 2016a; Shepard & Abuelsamid, 2016).

Cumulative sales already reached 2 million by the first quarter 2017 resulting in a 1.15% market penetration in new car sales in main markets (BNEF, 2017; BNEF, 2016a). In the leading Norwegian market, EV sales amounted to nearly 40% of new vehicle sales, with a fleet penetration of around 6.7% (OFV, 2017). This success is partly due to the favorable governmental policy schemes such as tax exemption of 25% VAT, free parking, access to bus lanes, in addition to high taxation on ICE cars (Figenbaum, Assum & Kolbenstvedt, 2015; IEA, 2016).

The increase in EV production and demand has intensified battery production and development, contributing to declining battery prices, dropping on average by 19% year-on-year since 2010.

This is mainly due to technology improvement, economies of scale, and market competition between manufacturers (BNEF, 2016b). As a result, the previously expensive EVs are becoming more affordable, and closer to parity with ICEs, as the battery cost of a battery electric vehicle constitutes about a third of the total production cost. In addition, EVs are now starting to enter the used car market where these vehicles have not previously existed. This implies that EVs will become a viable option for used car market buyers, as prices come to par with ICEs. However, uncertainties regarding the life of used batteries exist, and many consumers await maturity of the technology (BNEF, 2017).

1.2 Problematization

Political incentives, both fiscal and non-fiscal, together with technology development have played a vital role for consumers to buy EVs. However, surveys of EV owners show that social interaction also has significant impact on the purchase decision (Figenbaum et al., 2015; Figenbaum &

Kolbenstvedt, 2016; Kangur et al., 2017). Understanding the factors that affect the decisions and thereby the diffusion, is non-trivial as technology advancements, political measures and consumer preferences are continuously changing. Stakeholders have been uncertain of the technology shift

(22)

1 INTRODUCTION

order to gain insight on product investments and measures to deal with the change. Therefore, to create sufficient and concurrent insights there is a need to investigate the dynamics of the new and used car market, including how consumer actions and interactions impact the choice of vehicle and fuel type.

As diffusion of innovations involves complex phenomena and is defined as "the process by which an innovation is communicated through certain channels over time among members of a social system" (Rogers, 2003), there is an inherent need to simplify the process in order to understand and describe it. Therefore, to study these complex interactions between individuals, different diffusion models have been developed. One of these is the agent-based model approach, which captures single individuals’ needs and preferences and how they interact in the social system (Al- Alawi & Bradley, 2013). This allows for recreation of complex and emergent patterns of behavior that are not defined at the level of any individual, when data on aggregated behavior is limited (Adepetu et al., 2016). Therefore, the dynamics of the diffusion can be investigated by applying an agent-based model, creating insight into the market developments aiding stakeholders to make more informed decisions.

1.3 Purpose & Research Question

The purpose of this study is to investigate the diffusion of electric vehicles in the Norwegian market and assess the effectiveness of factors that affect the adoption rate. In order to address the purpose we aim to answer the following research questions:

1. How does an agent-based model describe the diffusion of electric vehicles?

(a) How do policies, technological, and social factors affect the diffusion of electric vehi- cles?

(b) What effects on the diffusion emerge when combining several factors simultaneously?

When investigating the diffusion factors and the different outcomes of their combination, insight into where the market is heading can be investigated. In doing so, another question arises for the operationalization of the purpose:

2. What can we learn about future developments in the Norwegian car market?

(a) What will the diffusion of electric vehicles look like until 2030?

(b) How is the used car market affected by the diffusion of electric vehicles?

1.4 Limitations & Delimitations

First of all, we delimit ourselves to the Norwegian market as the nature of the study requires a confined setting and consumer experiences with EVs. As the country with the highest EV penetration, there is comparatively much data available. However, EVs are in a relatively nascent state and sufficient data is sometimes limited. Hence, the study puts emphasis on analytical assessment of factors, rather than to yield quantitatively accurate predictions, which would be premature at this stage and out of the scope of the project. Nevertheless, the aim is to create as statistically accurate results as possible. Just as for many other volatile dynamic technologies, it is not easy to predict accurate sales and/or market penetration. However, the simulated results provide indications on future developments on which conclusion are drawn upon. For limitations regarding the model used in this paper, i.e. the STECCAR model, we refer to Kangur (2014), who describes the model’s limitations in depth.

2

(23)

1 INTRODUCTION

As the study regards diffusion of innovations and consumers behavior, we delimit the term electric vehicles to only include battery electric vehicles and plug-in hybrid vehicles. Hybrid vehicles without a plug-in feature are more similar to internal combustion engine vehicles, and are therefore omitted as they do not require the owner to charge the vehicle’s battery.

For the duration of this project, there have continuously been numerous developments in the area of electric transportation. These changes were accounted for at the highest level possible when analyzing literature and putting the results in context. However, in terms of access to data, vehicle data span years until the start of 2017 (start of the project). Naturally, any later incorporated data would require major rework and changes in the analysis, which the time frame of the project does not allow.

1.5 Contribution

Being one of the world’s largest industries, the automotive industry is under constant investigation both from politicians, the academic perspective, consulting firms, media as well as the public. Car sales have been closely monitored and projected on a global scale for an extended period of time, but EV adoption is in the context a relatively new phenomenon. Due to the rapidly changing environment, data and assumptions in literature quickly can become obsolete.

Hence, there is a gap in incorporating the latest trends in for example policy and technology, when trying to understand the main drivers for EV adoption. Furthermore, the connection to the used car market is of interest when deliberating if an EV is a viable purchase option.

In this study, we use an agent-based model to make an indication of the future development and assess factors that affect the adoption rate of EVs in the market. The model that is utilized was developed for the market in Netherlands, and modified it to fit the Norwegian consumer and market, our aim is to contribute to literature regarding agent-based modeling for diffusion of electric vehicles. Moreover, this study could provide valuable insight and implications of the diffusion of EVs for stakeholders in society, automakers, policy makers, and infrastructure partners such as energy companies providing the electricity for charging.

1.6 Thesis Structure

The remaining parts of the thesis are presented as follows: Chapter 2 provides an overview of the current electric vehicle market setting, including policy schemes, technological development and purchase behavior. In Chapter 3 the theoretical framing and literature review are presented.

Then follows Chapter 4, which describes the agent-based model and its components. Chapter 5 provides a description of the study’s methodology along with a discussion about the validity and reliability. In Chapter 6 the results derived from simulation runs in different scenarios are presented. The results are then analyzed in Chapter 7. The results and analysis are discussed and concluded in Chapter 8. Lastly, in Chapter 9 there is a discussion of further work and stakeholder implications.

(24)

(25)

2 Electric Vehicles

This chapter presents an overview of the current setting on electric vehicles, including policy support, technology development, and consumer purchase behavior. As this study mainly concerns Norway and the Norwegian setting, the focus of the chapter’s content is put there.

2.1 Overview

An electric vehicle (EV) uses an electric motor for propulsion, compared to internal combustion engine (ICE) cars that rely on fossil fuels to power the powertrain. The categorization of EVs constitute both battery electric vehicles (BEVs), and plug-in hybrid electric vehicles (PHEVs) (BNEF, 2016a). The former uses a battery for the electric engine; and the latter both have an electric and internal combustion engine. However, it is important to recognize that there is a difference between a PHEV and hybrid electric vehicle (HEV), which combines a conventional ICE propulsion system with an electric powertrain to achieve better fuel economy. Hence, it is not possible to charge and drive using only the electric engine in a HEV. As HEVs do not have a plug-in feature, we chose to omit them from this study and focus on BEVs and PHEVs as they share the all-electric drive and require the owner to charge the vehicle’s battery.

Many automakers are adding EVs to their product mix. At the Paris Auto Show in September 2016, the Chief Executive Officer for Daimler, Dieter Zetsche said: “We’re ready for the launch of an electric product offensive that will cover all vehicle segments, from the compact to the luxury class” (Nussbaum et al., 2016). The availability and performance of models are increasing, and as of today PHEVs have a battery capacity around 8 kWh, which translates to approximately 30-50 km of pure electric drive, only sufficient for shorter travels. BEVs on the other hand, have a battery capacity ranging between 30-100 kWh, allowing ranges between 200-500 km, enough for most daily travels. Nevertheless, there are more nuances to consumer choice than range and models. Figenbaum & Kolbenstvedt (2015b), identify two main dimensions affecting the diffusion of electric vehicles, i.e. Technology and Policy. Furthermore, the results in their study indicate that the Social aspect, i.e. consumer-to-consumer interactions, plays an imperative role for the purchase decision for alternative fuel vehicles.

2.2 Policy Support

The need for global reduction of carbon dioxide (CO2) emissions has over the previous years led to increased policies and incentives for “green transportation”. These policies include, but are not limited to, regulatory measures (e.g. emission regulations and fuel economy standards), financial levers (e.g. taxation on vehicles based on emission level (gCO2/km)) and other consumer direct incentives (IEA, 2016). The Norwegian Parliament has since 1990 provided policy support in various areas for an emergent EV diffusion. Compared to many other countries these incentives favor EVs over ICEs, where EV drivers enjoy several benefits such as free parking, access to bus lanes, exemption from taxes, toll fees (Figenbaum, Assum & Kolbenstvedt, 2015). Norway’s policies have led to an unmatched EV market share of new sales with 29% in 2016, moving towards 40% the first quarter of 2017 (EAFO, 2017). What is more, an interim goal is set by the National Transport Plan that all new passenger vehicles in Norway shall be emission free by 2025 leading up to the overall climate goal of carbon neutrality by 2030 (Norwegian Government, 2017).

(26)

2 ELECTRIC VEHICLES

2.2.1 Emission Standards

In the race of becoming the most environmentally sustainable and complying with regulations, automakers have put lots of efforts in combustion engine R&D in the strive for the lowest tailpipe emissions and fuel efficiency. The Volkswagen emission scandal of 2015 is an example of this competition, where VW went to great lengths to beat their competition, in this case fraudulently.

As BEVs have zero tailpipe emissions and are becoming more energy efficient, they have an advantage compared to ICEs in benefiting from regulations. However, it is only local pollution that is reduced in a shift towards EVs as the electricity needs to be produced elsewhere. For many countries, reaching climate change benefits is challenging due to a power generation that is dependent on coal, and thereby the whole EV vehicle life cycle needs to be assessed (IEA, 2016).

In Norway, on the other hand, about 96% of the electricity generation comes from renewable sources, why the positive environmental effects of EVs are conceded. The Norwegian goal is to be carbon neutral by 2030 (Stortinget, 2016), where the interim target of 2020 for new passenger vehicles is to have average emissions of 85 gCO2/km (Figenbaum et al., 2015) and as mentioned, the interim target of 2025 is that all new passenger vehicles will be emission free as proposed in the National Transportation Plan. This can be compared to the EU target of 95 gCO2/km for 2021 which represents a 40% reduction from the 2007 fleet average of 158.7 gCO2/km (European Commission, 2014), making Norway’s goal a comparative 46% reduction.

The interest in climate policy is strong as over recent years, Norwegian cities have not been able to comply with local air quality legislation leading up to a ban on diesel cars in the Oslo city centre on January 17th 2017 (see e.g. Bugge & Ridar, 2017). In addition to the temporary ban in Oslo, four of the largest cities in the world; Paris, Madrid, Athens and Mexico City will ban diesel cars and vans permanently by 2025 (Harvey, 2016), acting as frontrunners for other cities, paving the way for increased electrification of transports.

2.2.2 Fiscal Incentives

Taxes and subsidies have proven to be important in the diffusion of EVs. As mentioned, the sales figures of EVs in Norway are evidence of successful incentive schemes. Mock and Yang (2014) conclude that the total fiscal incentive in 2013 provided in Norway for BEVs were about 55% of the vehicle base price, and in the case of the Netherlands, the PHEV incentives were equivalent of about 75% of the vehicle base price. However, the authors also conclude that fiscal incentives are important, but it is not the only factor. In the United Kingdom the market share for both BEVs and PHEVs were low compared to the high fiscal incentive of 50% of the vehicle base price.

Figenbaum and Kolbenstvedt (2015a) state that the effective tax system in Norway along with the extended time frame for incentives might explain the discrepancy.

In relation to other countries, Norway’s vehicle taxes are high. Firstly, vehicles have a progressive registration fee where weight, engine power, CO2 and NOx emissions are taken into account.

Moreover, owning a vehicle implicates paying numerous taxes including; vehicle purchase tax VAT of 25%, fuel tax, annual registration tax, scrap deposit tax, income tax on company cars as well as road tolls (Norwegian Tax Administration, 2017). Measures to promote BEVs have been exemption from the registration tax, while smaller PHEVs often had no registration tax due to the low CO2 emissions. Today, EVs have a low annual registration tax, around 85% cheaper than ICEs (Norwegian Tax Administration, 2017). What is more, BEVs have an exemption from the VAT of 25% on the list price. Figenbaum and Kolbenstvedt (2015a) argue that the latter is an important fiscal incentive due to the fact that BEVs are and have been comparatively

6

(27)

2 ELECTRIC VEHICLES

expensive to produce, resulting in higher list prices. Hence, if the VAT exemption did not exist, BEVs would have a harder time competing with ICEs. Free toll roads is another incentive that have had large impact on the EV adoption in Norway. Costs can amount up to €2500/year for commuters, making BEVs favorable and emergent in cities as well as remote areas such as islands connected to the mainland with underwater tunnels (Figenbaum & Kolbenstvedt, 2015a).

However, the increase of EVs on toll roads have now led to the conclusion that they should contribute as well, and from March 2017 EVs in Oslo need to pay toll during rush hour of 10 NOK (58 NOK for diesel, 48 NOK for gasoline). Plans to raise fees from 2018 and make EVs pay at all times from 2020 are also on the table. Most likely, other cities will follow Oslo’s example and impose toll on EVs (Hegvik, Ertesvåg & Newth, 2016). Similarly to the toll, local municipalities can decide if EVs should be exempted from parking fees and have access to bus lanes. Before 2017, these initiatives were in effect nationwide and were according to Figenbaum and Kolbenstvedt (2015a) the most important incentives for EV uptake along with the VAT exemption.

2.3 Technological Development

Over recent years, the EV demand and production (along with other technology such as tablets, laptops and phones) has put more pressure on battery manufacturers to increase capacity and lower prices. Thanks to economies of scale, technology improvement, and market competition, battery prices per kWh have been dropping on average 19% year-on-year since 2010 (BNEF, 2016b). Notable in the strive for cheaper batteries with better capacity, is that several manufacturers plan to expand their current line or build new factories, such as Tesla’s “Gigafactory” in Nevada, U.S. As one of the leaders of EV production, Tesla supposedly will have a battery pack cost reduction by 30%, thereby enabling the lower cost of €32.500 for the Model 3 making Tesla more affordable in comparison to the Model S which is twice as expensive to the end customer (Dyer & Bryce, 2015). In addition to the U.S. factory, Tesla plans to build another Gigafactory in Europe and will compete with other manufacturers joining in the intensified electrification.

Northvolt is a newly founded company with former Tesla employees in the management who plan to build their own factory in Sweden or Finland in order to secure the European market and make it less dependent on the current Asian dominance in battery manufacturing (Milne, 2017).

As battery capacity and range of BEVs increase, the interest in building charging stations follow and the number of charging points in Norway are close to 9000, even though most EV owners charge their car at home (Nobil, 2017). To realize the possibility to take long-range trips above a single charge, incentives have been taken from both manufacturers and politicians. In Norway, charging stations have been subsidized from the government and the goal is to establish fast charging stations along the main roads every 50 km. On the automakers side, Tesla’s net of fast chargers are increasing and have up until 15th January 2017 been free of charge, but will ahead have a nominal fee for new Tesla owners in order to fund an expansion of the network (Tesla, 2016). In addition to conventional plug-in charging, more electric concept cars are equipped with wireless charging capabilities to relieve the customer of the cable in the future. Building charging networks is not all, as manufacturers try to reduce the perception that the battery is short-lived by offering warranties or utilizing lease plans for the battery thereby easing anxiety for vehicle life span as well as the secondhand value. For example, both Nissan and Tesla offer up to 8 years warranty on their batteries for capacity degradation (Nissan, 2017; Tesla, 2014).

(28)

2 ELECTRIC VEHICLES

In addition to technological advancements on the powertrain of the EVs, business model rein- vention, and autonomous driving are gaining more attention. Car sharing schemes are becoming more frequent and may have impact on EV adoption on a global level as they utilize a relatively high share of EVs, around 7% compared to the 1% in main markets (BNEF, 2016c). Customers who would not consider an EV for their personal use would then be exposed to them and form a better educated opinion. Moreover, the concept of shared economy is becoming more adaptable to people as it is realized by digitalization. The entrance of autonomous vehicles with great con- nectivity will perhaps allow for increased car sharing where vehicles constantly can be in motion instead of being parked (Gao et al., 2016). Recent developments of the autonomous technology shows that these vehicles might be on our streets in the near future, for example NIO plan to have its fully autonomous car in the U.S. in 2020 (NIO, 2017). However, there are many regulatory requirements that need to be resolved before fully autonomous vehicles are allowed, while the current advanced driving-assistance systems will have significant impact on how both consumers and regulators will be prepared for autonomous vehicles. A progressive scenario shows that fully autonomous cars could amount up to 15% of passenger vehicles sold globally in 2030 (Gao et al., 2016). In the analysis, this type of ownership and driving is left out due to the high uncertainty and chosen constraints in the model.

2.4 Purchase Behavior

Along with the global interest for environmentally sustainable solutions the demand for EVs has risen along with the number of models manufacturers supply. EV owners have various preferences about their vehicles but according to the most recent survey by Norwegian Institute of Transport Economics (TØI) by Figenbaum and Kolbenstvedt (2016), Norwegian owners share common motivation in the economy of use, the environmental factor and that the technology is future proof, where BEV owners also are motivated by incentives such as the free toll roads. Most of the EVs are bought new from a brand dealer constituting about 85% of sales, where a small portion of 4% BEV is peer-to-peer market and other markets are from used brand dealers as shown in Figure 1. Indirectly, this suggest a future consolidation towards a similar balance as for ICEs, presuming ownership structures will remain the same as the EV fleet grow. However, this might change with car sharing and autonomous cars; but as mentioned, due to the limitations of this study, this aspect is excluded and similar structures as of today are assumed.

8

(29)

2 ELECTRIC VEHICLES

Figure 1. Source of purchased vehicles for survey participants (Figenbaum & Kolbenstvedt, 2016) The surveys on Norwegian car owners showed that in addition to the location of purchase, a large influence on the decision to buy a BEV was word of mouth (Figenbaum, Kolbenstvedt &

Elvebakk, 2014; Figenbaum & Kolbenstvedt, 2016) why it is important to include network effects in the decision making process for EV purchases. Recommendations from the contact network of the BEV buyer had the most impact on the decision followed by information obtained from the brand or dealer. For PHEV and ICE the primary source of information was the dealer, with word of mouth as second for ICEs and third for PHEVs (Figenbaum & Kolbenstvedt, 2016).

2.4.1 The Used Car Market

The used car market in Norway experienced around 460 000 transactions in 2016 (OFV, 2017).

As seen in Figure 1, the market for used ICE vehicles is larger than the one for EVs. What they have in common is that vehicles often are sold at a price difference compared to a new vehicle due to depreciation. The depreciation rate is the difference between the list price of a new car and the resell price of the car after a period of time, i.e. the value difference. Many factors affect the depreciation such as: driven kilometers, maintenance costs, brand perception, vehicle features, and more. In addition, the depreciation is often one of the largest parts in the total cost of ownership (Hagman et al., 2016).

Although the market for used EVs is limited in Norway, the asking prices for the most popular models, the Tesla Model S, Nissan Leaf, and Mitshubishi Outlander PHEV have been strong.

However, the average BEV lose about 15% more of the value compared to ICEs (BNEF, 2017).

The general depreciation models used by financial institutes and developed for ICEs, set the expected depreciation rate at 50% after owning a car three years, and driving 45 000 km (Hagman et al., 2016). Between the two types of EVs, PHEVs and BEVs, generally retain more value due to concerns regarding battery life and range concerns which do not affect the PHEVs in the same sense. Even though consumers might be uncertain about purchasing used EVs, the Norwegian market is stable compared to other countries, for example the U.S. where residual values of EVs are significantly lower (BNEF, 2017).

(30)

(31)

3 Theoretical Framing and Literature Review

This chapter presents the theoretical background and literature review on which the study is based. First, general theory on diffusion of innovations which constitute the foundation for the subsequent theories is explained. Secondly, vehicle adoption models are evaluated and discussed.

Focus is subsequently put on the method of choice, namely agent-based models. The chapter is concluded with a reflection on previous research in the field, and a summary of agent-based diffusion models.

3.1 Diffusion of Innovations

Diffusion could be defined as “the process by which an innovation is communicated through certain channels over time among members of a social system” (Rogers, 2003). Communication referred to in this definition is a process in which individuals create and share information with each other, to attain a shared understanding of a subject. This implies the communication is a converging process as individuals consolidate toward a mutual interpretation ascribed to certain events, or the opposite, diverge and move further apart. Hence, this process implicitly makes complete diffusion unrealistic in the real world, as all individuals within the system would have to select the same behavioural option (Kangur, 2014). Therefore, the objective of a diffusion model is to indicate the adoption rate of an innovation, amongst a population, over time (Mahajan et al., 1991).

The heterogeneity of the population and the converging communication process, imply that not all individuals in a social system adopt to an innovation at the same time. Depending on characteristics, an individual’s propensity to adopt to a new technology is time dependent.

Rogers (1962) classifies different adopters into categories, based on when they first adopt to a technology. The innovativeness, or in other words, the degree to which an individual is prior to other members within the system to adopt to a new technology is what separates the different categories. The first two groups, i.e. Innovators and Early Adopters, constitute a relatively small part of the total market share and often have more to gain from the functionality of a new product, whereas the two subsequent groups, Early Majority and Late Majority, put more value on normative influences. It is first when these groups adopt to the new technology that it becomes the social norm, and self-sustaining. The last group to adopt is classified as Laggards.

It is important to acknowledge the various characteristics and preferences between the adopter groups, as different policies will be more or less effective during the diffusion process. In Figure 2, the normal distribution, i.e. bell-shaped frequency curve, represents number of individuals adopting per time unit; whereas the S-shaped sigmoid function, illustrates the same data on a cumulative basis.

(32)

3 THEORETICAL FRAMING AND LITERATURE REVIEW

Market share %

Innovators 2.5%

100

75

50

25

Early 0 Adopters

13.5%

Early Majority

34%

Late Majority

34%

Laggards 16%

Figure 2. Rogers diffusion of innovation. The bell-shaped frequency curve represent the number of individuals that adopt to a new technology per time unit, and the S-shaped curve is the cumulative basis (Rogers, 1962;2003)

As social phenomena involve complex interactions between individuals, that make decisions inde- pendently without any reference point; there is an inherent difficulty to understand and predict future outcomes such phenomena, e.g. vehicle diffusion. Therefore, researchers have developed models that try to simplify and illustrate this complex phenomenon. In the literature on diffusion models there are, in a broad sense, two categories of models; those that describe adopters as groups, rather than individuals; and models that account for the heterogeneity of the population (Geroski, 2000). The former builds on the premise that what is limiting the rate of adoption is the scarcity of information about the technology, how to use it, and what it does. The latter builds on the assumption that different parts of the population is likely to adopt to a technology at different times, due to the various attributes. To extract insights from the complex environment which characterize diffusion of innovations, models and computer simulations can provide environments which are appropriate to investigate the emergent processes.

3.2 Vehicle Adoption Models

Prior research assessing the market diffusion of EVs mainly constitutes three major modeling techniques; agent-based, consumer choice, along with diffusion and time series models (Al-Alawi

& Bradley, 2013). Most of these studies regarding the diffusion of EVs are based on diffusion and time series models, which focus on average driving patterns (see e.g. Becker et al., 2009;

Jeon, 2010; Lamberson, 2008; McManus & Senter, 2010; Struben & Sterman, 2008). These models assume a proportional adoption rate between the number of adopters and the remaining population, and are applied by calibrating Rogers’ market diffusion curve to new data or predict hypothetical growth rates (Gnann et al., 2015). However, as the individual driving and purchasing patterns constitute large heterogeneity these models tend to form inaccurate results (Smith et al., 2011; Gnann et al., 2015); especially as the procedure is sensitive in nascent states when limited amount of data is available (Gnann et al., 2015). Lately there has been an increase of

12

(33)

agent-based and consumer choice models, where the heterogeneity of the adopters is accounted for (see e.g. Sullivan et al., 2009; Neubauer et al., 2012; Gnann et al., 2015; Eppstein et al., 2011; Brown, 2013; de Haan et al., 2009; Mueller & de Haan, 2009; Cui et al., 2011; Kangur, 2014; Kangur et al., 2017). As contended by Axelrod & Tesfatsion (2005):

Understanding an economic system requires more than how individuals behave within the system, but also how interactions of many individuals leads to large-scale outcomes It requires more than to understand the individuals that comprise the system. It also requires understanding how the individuals interact with each other, and how the result can be more than the sum of the parts.

Furthermore, Gnann et al. (2014) among others, argue that agent-based models are more suited for complex and expensive purchase decisions. However, Al-Alawi and Bradley (2013) discuss result sensitivities on individual data which can have an impact if not assessed. Although, considering the complex economic system which characterizes the automotive market and the expensive nature of vehicle purchases, we thus intend to use an agent-based model for this study as it allows us to investigate emergent behavior that emanate from agent interactions, in a case where empirical data on an aggregated level is limited. Consequently, the literature review focuses on this type of diffusion model.

3.3 Agent-Based Modeling

Agent-based modeling (ABM) is a computational simulation method where a virtual environment is constructed in which actions and interaction between autonomous, proactive, reactive, and heterogenous agents are simulated (Al-Alawi & Bradley, 2013). This modeling technique is referred to by several common names, including: Agent-based modeling and simulation (ABMS), Agent-based Simulation (ABS), and Individual-based modeling (IBM). Nonetheless, the founding principles of the systems are the same. Agents are defined as entities or individuals that have control over the interactions, i.e. have the capability to make decisions, in the system. These decisions are either based on simple conditional rules (i.e. if-then statements) or by adaptive techniques, which assumes that certain situations and environments will affect the nature of different decision strategies, e.g. simple heuristics (Glöckner et al., 2014). Characteristics are defined for each agent which dictate their interactions among other agents and the environment, whereas the environment is defined with state variables that are representative for all agents within the model (Al-Alawi & Bradley, 2013). A representable structure of an agent-based model is depicted in Figure 3. The models’ bottom-up design enable the ABM to recreate and predict the appearance of a complex phenomena and emergent patterns of behavior that are not defined at the level of any individual agent (Adepetu et al., 2016). Further explanation of the model is described below.

(34)

Figure 3. General structure of an agent-based model (van Dam et al., 2012)

3.3.1 Agents

Agents, as the name suggest, are the main building blocks in agent-based models. They represent decision-making entities, which are dictated by certain rules and interact with each other and with the environment. The literature lacks a formal definition and consensus about what constitutes an agent. However, this characterization has been deemed useful for understanding what an agent is (Jennings, 2000): "An agent is an encapsulated computer system that is situated in some environment and that is capable of flexible, autonomous action in that environment in order to meet its design objectives." Wooldridge and Jennings (1995) provides an explanation in line with this definition by the following points:

(i) clearly identifiable problem solving entities with well-defined boundaries and interfaces;

(ii) situated (embedded) in a particular environment - they receive inputs related to the state of their environment through sensors and they act on the environment through effectors;

(iii) designed to fulfill a specific purpose - they have particular objectives (goals) to achieve;

(iv) autonomous - they have control both over their internal state and over their own behaviour;

(v) capable of exhibiting flexible problem solving behaviour in pursuit of their design objectives - they need to be both reactive (able to respond in a timely fashion to changes that occur in their environment) and reactive (able to act in anticipation of future goals)

Agent-based simulations can be made with single agents within an environment, but due to the inherent complexity in most social phenomena, simulations mostly require multiple agents;

to represent both the decentralized nature of the problem, and the multi-level and nonlinear interactions (Jennings, 2000). Following (iv), agents have control over its actions and are designed to fulfill an explicit purpose (iii). The interactions that arise between agents, in a multi-agent

14

(35)

simulation, could either be to achieve individual goals or to cope with the dependencies that emanate from the common environment. As seen in Figure 3, agents receive inputs from both the environment and other agents, which in turn affects the agent’s state, and subsequently its actions. It is from these interactions, and changes in behavior due to other agents that complex phenomenons emerge from otherwise simple deterministic rules.

As seen in Figure 3, an agent has a State that is defined by a collection of parameters and information about what the agent is at the moment. These states can either be internal, local or global. The internal state is what defines the specific agent, and comprise all agent properties (van Dam et al., 2012). An agent representing a consumer in a vehicle diffusion model could have internal state consisting of age, income, driving distance and current vehicle type etc. A simpler agent could be a light switch agent with the binary states on and off.

Furthermore, agents perform actions based on the decision rules derived from their current state.

In simplified terms, the agent in a vehicle diffusion model evaluate its needs and state, and subsequently form a decision to buy a new vehicle or not, based on a set of rules. For example,

"If current vehicle age >9 years, set state = buy new vehicle". It is from these state changes and actions, the agent behavior emerge as a result of the cumulative interactions between internal, local and global states and decision rules (van Dam et al., 2012).

3.3.2 Previous Research in Agent-Based Diffusion Models

There have been numerous studies using agent-based models to describe vehicle adoption: Ade- petu et al., 2016; Brown, 2013; Cui et al., 2011; de Haan et al., 2009; Eppstein et al., 2011; Gnann et al., 2015; Kangur et al., 2017; Plötz et al., 2014; Shafiei et al., 2012; Sullivan et al., 2009.

Eppstein et al. (2011) introduce a spatially explicit agent-based vehicle consumer model that measures the sensitivities and nonlinear interactions between various factors on PHEV market penetration. Each agent (consumer) is assigned specific attributes such as age, income, miles traveled per year, and typical years of ownership. Additionally, the model accounts for social effects (i.e. homophily and conformity) and media influence. The assigned “spatial neighbor- hood” works in conjunction with social networks to estimate agent network externalities in the model. However, the authors make numerous simplifying assumptions due to the nascent state of the EV market in 2011 and therefore assert they are not able to yield quantitatively accurate predictions. Despite the limitations of the model, qualitative insights into system behavior could be gained to assess which policies and procedures that may render most effective, and identify which data that may be important to examine. Still today, the EV market is in an early stage, however, more data is available to provide more accurate predictions and assumptions. This model serves as the basis for many subsequent papers within the field of vehicle adoption i.e.

Adepetu et al. (2016), Gnann et al. (2015), and Plötz et al. (2014).

Adepetu et al. (2016) develop a spatial agent-based model that aims to determine how different policies and battery technologies affect EV adoption, using San Francisco as a test city. This model comprises both vehicle adoption and usage, making it suitable for policymakers, utility operators, infrastructure providers, as well as EV manufacturers. However, certain limitations arise due to the rather compact geographical size of San Francisco, thus affecting the quantitative applicability of the study, why future research should intend to study a more representative geographical area that incorporates a larger variety of driving profiles. Additionally, the model presented by Adepetu et al. (2014) study is limited by a rather simple total cost of ownership (TCO) estimation process, and does not account for long-distance trips in the decision process.

(36)

Knox County, Tennessee, similar to Adepetu et al. (2016). However, it focuses more on charging infrastructure rather than EV adoption in the different scenarios.

More comprehensive TCO estimations and driving profile data are used by Gnann et al. (2015) and Plötz et al. (2014) in their studies on EV market evolution in Germany until 2020. Driving profiles are based on individual measurements, and differentiated into three different user groups;

i.e. users of private, fleet vehicle, and company cars, due to noteworthy differences in driving behavior for the different profiles. The authors assert that distribution and regularity of trip lengths vary greatly between agents, which have a large influence of the TCO and subsequently the potential use of EVs over conventional vehicles. Although, uncertainties in future exogenous developments, i.e. oil and battery prices and the inherent challenge to estimate fuel usage and cumulative driving distance makes it difficult for consumers to calculate and weigh the environmental and financial trade-offs between different vehicle alternatives (Eppstein et al., 2011). This lack of knowledge by consumers is supported by several consumer purchase studies regarding alternative fuel vehicle purchases which asserts that consumers tend to make decisions on non-financial reasons (i.e. image, “perceived greenness” etc.), rather than rational reasoning on e.g. total cost of ownership and fuel economy (Heffner et al., 2007; Turrentine & Kurani, 2007). However, despite the tendency of non-rational reasoning several studies illustrate the weight of fiscal incentives (directly affecting the TCO), play an imperative role for diffusion of EVs (see e.g. Figenbaum & Kolbenstvedt, 2015b; Gnann et al., 2015; Adepetu et al., 2016;

Struben & Sterman, 2009). This is especially important as EVs historically have not been in parity with ICEs regarding performance and functionality, why incorporation of both rational, i.e. financial, and non-financial decision reasoning, i.e. social aspects, is important to investigate.

A study that incorporates a more comprehensive decision process is Kangur (2014) that models the EV penetration in the Dutch market using a spatially explicit ABM, which builds on the

“Consumat framework” developed by Jager (2000). The Consumat framework captures the main behavioral principles of consumer decision making by a collection of psychological meta-models.

As contended in the study, this approach enables inclusion of more complex behavioral rules in a multidisciplinary context, where technology development, policies, and behavioral effects are studied simultaneously (Kangur, 2014; Kangur et al. 2017). The agent-model, named STECCAR short for ‘Simulating the Transition to Electric Cars using the Consumat Agent Rationale’, in which the Consumat framework is applied, comprises an elaborate process that explores scenarios and diffusion patterns depending on different technology developments and policy stringencies.

This model constitutes the most comprehensive and elaborate process we have found in the literature and is well aligned with the limitations and domain of our study. Furthermore, as both the Consumat framework (see e.g. Acosta-Michlik & Espaldon, 2008; Jager et al., 2001;

Brouwers & Verhagen, 2003) and the STECCAR model (Kangur, 2014; Kangur et al., 2017) have been successfully implemented in several different consumer domains, and most recently for vehicle purchase decision. Therefore, we contend it serves our purpose well and thus chose to adopt it and use it as the basis for this study.

3.3.3 Summary Agent-Based Modeling

In summary, agent-based models facilitate the ability to capture complex structures and dynamics, in the absence of the knowledge of global interdependencies; meaning that you may know very little about the effects on an aggregate level, but have a perception of how individuals of the process behave (Borshchev & Filippov, 2004). This allows investigation of emerging global behavior from simple deterministic rules. The main advantage with an ABM is that it accounts for the heterogeneity of the population and thus capture emergent phenomena and provide a

16

(37)

natural description of the system (Bonabeau, 2002). It also facilitates a degree of flexibility, where e.g. agent complexity can be modified, or the level of aggregation i.e. subgroups or individuals. However, the complexity and heterogeneity are twofold; they also constitute the main disadvantages with the modeling technique, as it becomes harder to validate and verify.

Furthermore, individual data and elasticities can have large impact on the results if sensitivities are not assessed (Al-Alawi & Bradley, 2013). Although considering the inherent complexity of socio-economic phenomena, it is rather a characteristic of the socio-economic systems than the model itself. Even though there are challenges with agent-based models, it is increasingly being realized in the literature, that the social world needs to be looked upon as a complex adaptive system, where interactions between entities, i.e. consumers, in this case, are multi-level and nonlinear (van Dam et al., 2012).

(38)

(39)

4 Model Description

While Chapter 3 provided a brief overview and a review of agent-based modeling, this chapter describes the underlying the chosen agent-based model, STECCAR, and the framework it is built upon. First, the framework is presented, then the model, and last how the data was parameterized to fit the model.

4.1 Consumat Framework

Purchase decisions are often made by influences from social networks. Consumers may also rely on habits, making comparisons, imitating others or using their social network to get advice. In the STECCAR model, the Consumat framework is applied to these aspects of purchase decisions.

The framework, depicted in Figure 4, aims at connecting the decision making behaviors by accounting for the needs and abilities of agents, including when an agent switches their source of decision strategy. The main components of the revised Consumat framework (Consumat II) suggested by Jager and Janssen (2012) are presented here. For further reading, we refer to Jager (2000), and Jager and Janssen (2012).

Figure 4. Consumat framework overview (Jager, 2000)

4.1.1 Behavior Drivers

The drivers for behavior in the model are needs, that when fulfilled result in satisfaction. In this case, the needs are satisfied by the purchase of a new vehicle. The needs can be divided into three behavior-driving forces: existence, social belonging and personal preferences. Existence refers to the acts by agents of having means of income, food, or housing. In order to avoid loss of existence, agents act to renew their resources. Social belonging refers to interactions in the agent’s network, group affiliation, and social status. Personal preferences refers to satisfying the subjective taste and liking of an individual. Differences between agents are the balancing

(40)

4 MODEL DESCRIPTION

influences of their social network, i.e. other agents. The behavior options that agents perform relate to the abilities of an agent, i.e. the actual capacity to carry out an action. As agents carry out choices, the behavioral opportunity is connected to the abilities where for example resources can be used when choosing to buy a new vehicle. In addition, agents possess a memory where previously made choices are affecting new states of behavior and abilities.

4.1.2 Decision Strategy

The Consumat framework has a simple structure for decision making, which is an important element in the approach. The satisfaction level of the agent indicates the success of previous choices, so if an agent has a high degree of satisfaction, the need for decision making at the moment is low. However, when an agent is dissatisfied, the urge to make decisions becomes higher in order to increase satisfaction. Moreover, when there are numerous choices to decide from, one might be uncertain of the outcome of the decision. By looking at other people’s experience and their behavior, the uncertainty can be eased. Most often, these interactions occur when people share similar attitudes, values etc. Likewise, the Consumat framework accounts for similarity by basing chances of interaction on similarity, and thereby a social network can be constructed.

As agents’ behavior change over time, the dynamic network will follow. The level of satisfaction and uncertainty of the agent corresponds to the choosing of one of the four "cognitive processing"

strategies, as shown in Figure 4.

When satisfaction is high and uncertainty low, repetition occurs, which drives habitual behavior.

If, however, uncertainty would be high while satisfaction remains high, agents engage in imitation.

Low satisfaction requires agents to put more effort in bettering their state. Therefore, when agents are certain but not satisfied they will assess viable options and start optimizing and thereby maximizing the utility of a choice. If agents are uncertain while having low satisfaction, inquiring occurs where the agent will compare behavior to similar others and copy it if the expected outcome increases satisfaction.

4.1.3 Memory

The experiences from choices combined with behavior make up the memory of the agent. In- formation gathered from inquiring and optimizing is stored in the memory and used next time agents make decisions. Hence, the memory is updated only when agents are engaged in optimizing or inquiring, making it possible for a satisfied agent to stay in repetition without updating its information on better opportunities. By combining status of capacity and requirements for consuming a certain opportunity, the behavioral control is formalized in the memory, meaning the agent knows for example if it can financially afford a new vehicle. The individual agent’s behavior is collectively aggregated from the impacts affected by the environment, and/or other agents.

4.1.4 Framework Evaluation

The Consumat framework has previously been successfully applied in various domains, e.g. adap- tion to climate change by farmers, flood management, household dynamics, and most recently in the STECCAR (Kangur, 2014; Kangur et al., 2017). Hence, we know that it is a useful tool for modeling diffusion affected by social processes. However, Kangur (2014) argues that one con- cern regarding the framework is that needs are not ordered but balanced in line with a personal weighing function. The result would be, however unlikely, that symbolic and financial aspects outweigh the car’s ability to meet the driver’s behavior. If so, a person would buy a vehicle for

20

(41)

4 MODEL DESCRIPTION

environmental or financial reasons, even if the vehicle does not meet the daily travel distances.

Therefore, the assumption is that being able to travel the required distance is a basic need and failure to do so will result in a penalty in the overall satisfaction level of the agent. Moreover, another assumption accounted for is the influence from social networks regarding the purchase of a vehicle. As it is a substantial investment, it is unlikely that an individual would replace his or her car to a model similar to a peer as soon as uncertainty arises. Hence, the heuristic-based decision strategies can be viewed more as information seeking strategies (Kangur, 2014).

4.2 The STECCAR Model

The previous section presented the Consumat framework II on which the STECCAR model is built. This section will review how the model is constructed using the Consumat framework. The model that we utilize was developed by Kangur (2014), which we refer to for a more in-depth description.

4.2.1 Overview

The STECCAR model builds on a set of agents that own a vehicle that satisfies their personal needs. The model is written in Java and is executed with Repast Simphony, an open source agent- based modeling platform. The original model was parameterized using data from a nationwide survey in the Netherlands in June 2012, similar to the COMPETT survey (see Figenbaum, Kolbenstvedt, & Elvebakk, 2014; Figenbaum & Kolbenstvedt, 2015b) in Norway we use in this study. Both surveys capture data including individual characteristics, vehicle model, type of ownership, driving behavior and perception of EVs and the likeliness to purchase an EV in future. Necessary modifications have been made to the original code to fit the Norwegian setting and the market conditions.

In the model, there are three types of fuel technologies available: internal combustion engine (ICEs), battery electric vehicles (BEVs), and plug-in hybrid electric vehicles (PHEV). An agent’s behavior is restricted by its financial state and the refueling ability of the different technologies.

What Drives Electric Vehicle Diffusion?: An Agent-Based Approach to Assess Factors and Market Effects in Norway

What Drives Electric Vehicle Diffusion?

An Agent-Based Approach to Assess Factors and Market Effects in Norway

ALBIN GROPP

FREDRIK OHLSSON

What Drives Electric Vehicle Diffusion?

An Agent-Based Approach to Assess Factors and Market Effects in Norway

Albin Gropp

Fredrik Ohlsson

Vad påverkar diffusionen av elektriska fordon?

En agentbaserad modell för att bedöma faktorer och marknadseffekter i Norge

Albin Gropp Fredrik Ohlsson

Abstract

Sammanfattning

Contents

List of Figures

List of Tables

Acronyms

Foreword and Acknowledgements

1 Introduction

1.1 Background

1.2 Problematization

1.3 Purpose & Research Question

1.4 Limitations & Delimitations

1.5 Contribution

1.6 Thesis Structure

2 Electric Vehicles

2.1 Overview

2.2 Policy Support

2.3 Technological Development

2.4 Purchase Behavior

3 Theoretical Framing and Literature Review

3.1 Diffusion of Innovations

3.2 Vehicle Adoption Models

3.3 Agent-Based Modeling

4 Model Description

4.1 Consumat Framework

4.2 The STECCAR Model